An EMD-based invariant feature extraction algorithm for rotor bar condition monitoring

Development of portable devices for reliable condition monitoring of induction machines has become the goal of many researchers. In this context, the development of robust algorithms for the automatic diagnosis of electromechanical failures plays a crucial role. The conventional tool for the diagnostic of most faults is based on the FFT of the steady-state current. However, it implies significant drawbacks in industrial applications in which the machine does not operate under ideal stationary conditions (e.g. presence of pulsating load torques, supply unbalances, noises...). In order to overcome some of these problems, a novel transient-based methodology (Transient Motor Current Signature Analysis, TMCSA) has been recently proposed. The idea is to analyze the current demanded by the machine under transient operation (e.g. during the startup) by using proper Time Frequency Decomposition (TFD) tools in order to identify the presence of specific patterns in the time-frequency map caused by the characteristic evolutions of fault-related components. However, despite the excellent results hitherto obtained, the qualitative identification of the patterns requires a certain user expertness, which implies difficulties for the automation of the diagnosis. A new algorithm for the automatic diagnostic of rotor bar failures is proposed in this paper. It is based on the application of the Hilbert-Huang Transform, sustained on the Empirical Mode Decomposition process, for feature extraction, and the further application of the Scale Transform (ST) for invariant feature selection. The results prove the reliability of the algorithm and its generality to automatically diagnose the fault in machines with rather different sizes and load conditions.